A Novel Tool for Assessing the Impact of Diagnostic Options for Tuberculosis Control Using System Dynamics

Abstract

Tuberculosis (TB) is the single largest infectious cause of death among young people and adults in the world, accounting for nearly two million deaths a year. Despite being curable TB will continue to be a major problem in the foreseeable future. Eighty percent of TB cases are carried by just 22 countries since poverty and lack of access to health care exacerbate the TB problem. The three main challenges for TB control are case detection, diagnosis and adherence to medication regimes. Of these, diagnosis of TB is one of the most pressing challenges that was identified through working closely with a TB program in rural Bihar, India. Inaccurate diagnostic tests, lack of knowledge from medical practitioners and delays in obtaining diagnostic results are some of the major issues faced in diagnosing TB. The diagnostic problem arises because a case of TB is not always obvious, and does not always show up on a single test, this means that the status of a patient with TB symptoms can be uncertain. In light of the uncertainty involved in the types of patients who present to a TB program (e.g. severe TB, mild TB and not TB), trying to find a middle ground between over diagnosis (falsely diagnosing TB) and under diagnosis (falsely diagnosing not TB) is the challenged faced when designing a diagnostic program. In order to combat diagnostic inaccuracies and delays, there are several novel diagnostic tools currently in development but there is no robust method for evaluating their potential impact. Hence there is a great need for a robust tool which is able to evaluate new tools in the form of performance metrics including accuracy, costs and speed. This tool should be adaptable to several contexts, provide specific performance metrics and have the ability to guide policy makers when making recommendations as to the introduction of new diagnostic interventions. The approach taken to tackle this problem was two-fold. Firstly a collaboration was formed with a TB program operating in rural Bihar, India. This program was used as a case study on which to base the model. Through field work, an understanding of the overall system was gained as well as focused understanding of challenges in diagnostics. Primary data generated through the operation of the program was used to form an understanding of the demographics of the area and treatment seeking behavior, as well as typical practices of private providers. Secondly in order to deliver the needed tool for policy makers, a mathematical model was developed which represents the diagnostic process of a generic TB program. Within this model, the diagnostic tools and algorithms can be adjusted and redesigned to test the impacts of certain changes on diagnostic success rates, costs and delays. By running simulations in the model the tradeoffs in performance metrics were observed for different diagnostic tools and algorithms. The result of these simulations is a set of sensitivity analyses which show the impact of making improvements on diagnostic tools or indeed replacing them with new tools. We are able to put a value on marginal improvements to diagnostic accuracy, and to suggest new algorithms for TB diagnosis. Additionally the existence of a large informal network of unlicensed medical practitioners adds to the challenge of TB control as these private providers often prescribe inappropriate treatment and can even exacerbate the epidemic. However they are a potential resource to be tapped in order to improve case finding, since these practitioners often have strong social skills and are the first point of call for many treatment seekers. The tool can be extended to investigate the effects of engaging with private providers to enhance the referral of TB suspects to a TB program, and groundwork has been laid for further work in this area. This study is unique as it provides the first comprehensive tool for evaluating diagnostic options, with the added capability of looking at dynamic properties of a TB program, such as robustness to surges in patient referrals, seasonal effects and the feedback loops which govern disease transmission. To the author's knowledge there is no other tool with such capabilities. The model produced in this study will enable the simulation of interventions without the need to test them on patients in the field or to invest in potentially ineffective technologies.

Date of Award	Dec 2011
Original language	American English
Supervisor	Scott Kennedy (Supervisor)